Falcon3-10B-Base: benchmark results
TII's Falcon3-10B-Base, a raw pretrained model depth-upscaled from Falcon3-7B-Base via continued pretraining on 2 trillion tokens, supporting English, French, Spanish and Portuguese with 32K context. Provider: TII. Released 2024-12-17. Access: Open.
Aggregate Rating 1406 ± 19, rank #1206 of 1656 rated models, from 210 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Swallow - Pre-trained English - GSM8K | 80.2 | Exact Match (%) | 87.8 |
| Open Portuguese LLM - ENEM | 74.32 | Accuracy (%) | 86.7 |
| Open Portuguese LLM - OAB Exams | 54.58 | Accuracy (%) | 86.7 |
| Open LLM Leaderboard - GPQA | 34.56 | Score | 86.3 |
| Open Portuguese LLM - BLUEX | 64.53 | Accuracy (%) | 85 |
| Open LLM Leaderboard - BBH | 59.5 | Score | 81 |
| Swallow - Pre-trained English - BBH | 76.2 | Exact Match (%) | 80.6 |
| Open LLM Leaderboard - MATH Level 5 | 24.92 | Score | 78 |
| Open LLM Leaderboard - MuSR | 43.98 | Score | 77.9 |
| Swallow - Pre-trained English - HumanEval | 54.3 | Pass@1 (%) | 77.6 |
| Swallow - Pre-trained English - MATH | 49.2 | Exact Match (%) | 77.6 |
| Open Japanese LLM - Wiki PAS SET F1 | 8.11 | Score (%) | 77.2 |
Interactive version: theaggregate.ai/model?slug=falcon3-10b-base · How It Works · Data refreshed daily, snapshot 2026-10-20.